Model card summary
Model: turnout propensity v1. Type: deterministic feature-weighted heuristic. Status: available for ranking and segment comparison. Owner: VoterFile. Review date: July 11, 2026.
The intended use is prioritizing field, mail, vote-by-mail follow-up, and volunteer capacity within an authorized voter universe. The score must not be represented as candidate support, ballot choice, or a verified future action.
- Output: integer score from 0 to 100
- Primary signal: participation count normalized to available elections
- Secondary signals: age band, vote-by-mail status, registration status, and registration recency
- No protected-class input and no candidate-support target
What the score is for
Turnout propensity is a ranking signal. It helps campaigns decide which voters are likely to vote, which voters need turnout support, and which universes are realistic for canvassing, mail, SMS, or volunteer calls.
The score is not a guarantee of future behavior. It is a way to order work based on prior participation and registration signals that campaigns can review.
Signals and exact v1 weights
Past participation contributes up to 60 points, normalized by the number of elections available in that dataset. Age contributes 0–20 points through published bands, permanent vote-by-mail adds 10, inactive registration subtracts 20, and registration recency contributes minus 5 for less than one year or plus 10 otherwise. The final score is rounded and clamped to 0–100.
California's shared file uses a 10-election denominator. Bring-your-own workspaces calibrate the participation term to the number of distinct elections present in that workspace's history; a file with no usable history receives no participation points.
- Past participation count
- Age-based turnout curve
- Permanent vote-by-mail signal
- Active or inactive registration status
- Registration recency
How campaigns can use it
High-propensity voters can help a campaign define reliable supporters, likely mail voters, or persuasion targets. Mid-propensity and drop-off voters can support turnout programs when the campaign has the capacity to follow up.
The best use is comparative: inspect a district, compare segment sizes, then export only the universe that matches the campaign's time, budget, and volunteer capacity.
Validation status
The v1 score has unit and boundary tests for formula behavior, but VoterFile has not published a prospective accuracy, calibration, lift, or subgroup-fairness study on a held-out election. It should therefore be treated as an ordering heuristic rather than a probability such as '72% likely to vote.'
A future model should be trained and evaluated on time-separated election history, report calibration and lift by geography and demographic group, compare against a participation-only baseline, and publish the exact evaluation window before replacing v1.
Known limitations and safe use
Voting history records participation, not ballot choice. Past turnout can encode unequal access, campaign attention, mobility, and incomplete history. Age or registration dates may be missing, and election-history depth varies by state and source file.
Campaigns should compare multiple score bands, inspect counts and geography, preserve a path for organizer judgment, and avoid using the score as the sole basis for denying information or contact to a voter.
- Not a calibrated probability
- Not a candidate-support or persuasion score
- Sensitive to missing or shallow election history
- Can reproduce historical turnout disparities
- Requires campaign judgment and periodic re-review
Complete California access review before any file upload.
Submit the campaign, requester, issuing source, viewer list, geography, and intended use. VoterFile does not open upload, import, browse, export, or voter-level modeled output until the California activation checklist is complete.
Questions
Is turnout propensity the same as support score?
No. Turnout propensity estimates likelihood to participate. It does not estimate whether a voter supports a candidate or ballot measure.
Does voting history show who someone voted for?
No. California voter participation history can show whether a voter participated in an election and the voting method, but it does not include ballot choices.
Is a score of 72 a 72% chance of voting?
No. The current v1 score is a bounded ranking heuristic, not a calibrated probability. It is useful for comparing voters or segments, not for making a literal percentage claim.
Has the model been independently validated?
Not yet. The formula has automated behavior tests, but VoterFile has not published an independent held-out-election accuracy, calibration, or subgroup-fairness study.